drivers: net: xgene: fix Tx flow control

Currently the Tx flow control is based on reading the hardware state,
which is not accurate since it may not reflect the descriptors that
are not yet reached the memory.

To accurately control the Tx flow, changing it to be software based.
Signed-off-by: NIyappan Subramanian <isubramanian@apm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

drivers/net/ethernet/apm/xgene/xgene_enet_main.c

@@ -289,6 +289,7 @@ static int xgene_enet_setup_tx_desc(struct xgene_enet_desc_ring *tx_ring,
 				    struct sk_buff *skb)
 {
 	struct device *dev = ndev_to_dev(tx_ring->ndev);
+	struct xgene_enet_pdata *pdata = netdev_priv(tx_ring->ndev);
 	struct xgene_enet_raw_desc *raw_desc;
 	__le64 *exp_desc = NULL, *exp_bufs = NULL;
 	dma_addr_t dma_addr, pbuf_addr, *frag_dma_addr;
@@ -419,6 +420,7 @@ static int xgene_enet_setup_tx_desc(struct xgene_enet_desc_ring *tx_ring,
 	raw_desc->m0 = cpu_to_le64(SET_VAL(LL, ll) | SET_VAL(NV, nv) |
 				   SET_VAL(USERINFO, tx_ring->tail));
 	tx_ring->cp_ring->cp_skb[tx_ring->tail] = skb;
+	pdata->tx_level += count;
 	tx_ring->tail = tail;
 
 	return count;
@@ -429,14 +431,13 @@ static netdev_tx_t xgene_enet_start_xmit(struct sk_buff *skb,
 {
 	struct xgene_enet_pdata *pdata = netdev_priv(ndev);
 	struct xgene_enet_desc_ring *tx_ring = pdata->tx_ring;
-	struct xgene_enet_desc_ring *cp_ring = tx_ring->cp_ring;
-	u32 tx_level, cq_level;
+	u32 tx_level = pdata->tx_level;
 	int count;
 
-	tx_level = pdata->ring_ops->len(tx_ring);
-	cq_level = pdata->ring_ops->len(cp_ring);
-	if (unlikely(tx_level > pdata->tx_qcnt_hi ||
-		     cq_level > pdata->cp_qcnt_hi)) {
+	if (tx_level < pdata->txc_level)
+		tx_level += ((typeof(pdata->tx_level))~0U);
+
+	if ((tx_level - pdata->txc_level) > pdata->tx_qcnt_hi) {
 		netif_stop_queue(ndev);
 		return NETDEV_TX_BUSY;
 	}
@@ -539,10 +540,13 @@ static int xgene_enet_process_ring(struct xgene_enet_desc_ring *ring,
 	struct xgene_enet_raw_desc *raw_desc, *exp_desc;
 	u16 head = ring->head;
 	u16 slots = ring->slots - 1;
-	int ret, count = 0, processed = 0;
+	int ret, desc_count, count = 0, processed = 0;
+	bool is_completion;
 
 	do {
 		raw_desc = &ring->raw_desc[head];
+		desc_count = 0;
+		is_completion = false;
 		exp_desc = NULL;
 		if (unlikely(xgene_enet_is_desc_slot_empty(raw_desc)))
 			break;
@@ -559,18 +563,24 @@ static int xgene_enet_process_ring(struct xgene_enet_desc_ring *ring,
 			}
 			dma_rmb();
 			count++;
+			desc_count++;
 		}
-		if (is_rx_desc(raw_desc))
+		if (is_rx_desc(raw_desc)) {
 			ret = xgene_enet_rx_frame(ring, raw_desc);
-		else
+		} else {
 			ret = xgene_enet_tx_completion(ring, raw_desc);
+			is_completion = true;
+		}
 		xgene_enet_mark_desc_slot_empty(raw_desc);
 		if (exp_desc)
 			xgene_enet_mark_desc_slot_empty(exp_desc);
 
 		head = (head + 1) & slots;
 		count++;
+		desc_count++;
 		processed++;
+		if (is_completion)
+			pdata->txc_level += desc_count;
 
 		if (ret)
 			break;
@@ -580,10 +590,8 @@ static int xgene_enet_process_ring(struct xgene_enet_desc_ring *ring,
 		pdata->ring_ops->wr_cmd(ring, -count);
 		ring->head = head;
 
-		if (netif_queue_stopped(ring->ndev)) {
-			if (pdata->ring_ops->len(ring) < pdata->cp_qcnt_low)
-				netif_wake_queue(ring->ndev);
-		}
+		if (netif_queue_stopped(ring->ndev))
+			netif_start_queue(ring->ndev);
 	}
 
 	return processed;
@@ -1033,9 +1041,7 @@ static int xgene_enet_create_desc_rings(struct net_device *ndev)
 	pdata->tx_ring->cp_ring = cp_ring;
 	pdata->tx_ring->dst_ring_num = xgene_enet_dst_ring_num(cp_ring);
 
-	pdata->tx_qcnt_hi = pdata->tx_ring->slots / 2;
-	pdata->cp_qcnt_hi = pdata->rx_ring->slots / 2;
-	pdata->cp_qcnt_low = pdata->cp_qcnt_hi / 2;
+	pdata->tx_qcnt_hi = pdata->tx_ring->slots - 128;
 
 	return 0;
 

drivers/net/ethernet/apm/xgene/xgene_enet_main.h

@@ -155,11 +155,11 @@ struct xgene_enet_pdata {
 	enum xgene_enet_id enet_id;
 	struct xgene_enet_desc_ring *tx_ring;
 	struct xgene_enet_desc_ring *rx_ring;
+	u16 tx_level;
+	u16 txc_level;
 	char *dev_name;
 	u32 rx_buff_cnt;
 	u32 tx_qcnt_hi;
-	u32 cp_qcnt_hi;
-	u32 cp_qcnt_low;
 	u32 rx_irq;
 	u32 txc_irq;
 	u8 cq_cnt;